Abstract
Dehazing is an emerging computer vision research area aiming to restore image visibility by eliminating haze, a degradation caused by atmospheric scattering and environmental pollution. Accurate modeling of hazy images is challenging, despite the widespread use of the atmospheric scattering model. However, single image dehazing is more difficult since it requires an accurate measurement of the ambient light and transmission map. This work proposes a multi-scale Gradient domain Weighted Guided Filter (GWGIF) based dehazing method for hazy photos and videos. This work has established a computationally efficient method for estimation of physical model parameters. Initially, an image pyramid was constructed from the hazy input image. Subsequently, at the coarsest level of the pyramid, the scene transmission map and atmospheric light were estimated. Following that, with the help of GWGIF, transmission at the level of its finest has been obtained. Transmission map estimation has been done using Minimum Preserving Subsampling (MPS) and then by iterative up sampling with GWGIF has been applied to prevent information loss. Thereafter, Gradient Based Correlation Factor (GCF) has been introduced to expand the usage of the single-image dehazing technology to live video dehazing, thereby reducing dehazed videos flickering artifacts.